LLM Embeddings for Deep Learning on Tabular Data

Boshko Koloski Andrei Margeloiu Xiangjian Jiang, Blaž Škrlj Nikola Simidjievski Mateja Jamnik · 2025

Read LLM Embeddings for Deep Learning on Tabular Data by Boshko Koloski Andrei Margeloiu Xiangjian Jiang, Blaž Škrlj Nikola Simidjievski Mateja Jamnik (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2502-11596.txt.

Description

Boshko Koloski and co-authors turned tables into text for AI in two thousand twenty-five. Deep learning on tables needs numbers and categories embedded well. Standard encoders treat each data type separately. That blocks transfer of knowledge across tables. The team instead rewrites each table row as text. A pretrained language model then embeds the text.

The embeddings plug into existing table networks unchanged. Seven classification datasets tested the idea. Accuracy beat strong baselines like tuned transformers. Gains held across several network backbones. Text form lets models use word knowledge about categories. It also unifies handling of mixed column types. The trick is simple enough for wide adoption in analytics.

Catalog record

CREATORBoshko Koloski Andrei Margeloiu Xiangjian Jiang, Blaž Škrlj Nikola Simidjievski Mateja Jamnik
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE29 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2502-11596

Browse the full library · About the Tower of Babel